Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Negative Variance Inflation Factor-collin command

From   "Diego Rei" <>
To   <>
Subject   Re: st: Negative Variance Inflation Factor-collin command
Date   Tue, 22 Feb 2005 18:53:41 +0100

Thank you Richard for your help.
After some more tries I think that the problem is really the number of
variables used.
Of course it is impossible to get a negative vif, being this defined as
1-r2, however,
 because of the way the vif is constructed (regression of independent
variables on the others) I think you are right that the original command
used in the regression does not change the results, because the
accessory regressions should still be done by Stata through the reg
Best Regards,

>>> 02/22/05 5:30 pm >>>
At 04:56 PM 2/22/2005 +0100, Diego Rei wrote:
>Good day to all,
>I am trying to run a  collinearity test using the command "collin",
>panel data with several dummies and I get a very large negative value
>the mean vif, apparently due to these dummies, which have each a
>negative value. However, if I run the simple vif command (which I
>use after xtgls, the command I d rather use) I get very sensible
>positive values.

I assume you gave a -reg- command followed by -vif-?  If so, I am
-collin- isn't producing the same results (and even more surprised that
are getting negative vifs!  That isn't supposed to be possible, is

A few things to try:

* Make sure you have the most current version of -collin-.  It has been

updated a few times over the past year.

* Make sure sample selection is the same; missing data on y can cause 
-collin- to give different results than a corresponding -reg- -vif- 
combo.  E.g. do something like

collin x1 x2 x3 if !missing(y)

* Perhaps there is some bug or limitation in -collin- that zaps it when

there are too many variables?  It uses all these matrix algebra
so I am not sure what its limits are.

* If all you want are the vifs, there is nothing particularly wrong
what you are doing.  Just remember that whatever you do, this is
more like an approximation than an exact result, since the time series

nature of the data is not being taken into account.

Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
FAX:    (574)288-4373
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu 
WWW (personal): 
WWW (department): 

*   For searches and help try:
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index